Invariant Feature Research Articles

Multi‐View Self‐Supervised Auxiliary Task for Few‐Shot Remote Sensing Classification

ABSTRACTIn the past few years, the swift advancement of remote sensing technology has greatly promoted its widespread application in the agricultural field. For example, remote sensing technology is used to monitor the planting area and growth status of crops, classify crops, and detect agricultural disasters. In these applications, the accuracy of image classification is of great significance in improving the efficiency and sustainability of agricultural production. However, many of the existing studies primarily rely on contrastive self‐supervised learning methods, which come with certain limitations such as complex data construction and a bias towards invariant features. To address these issues, additional techniques like knowledge distillation are often employed to optimize the learned features. In this article, we propose a novel approach to enhance feature acquisition specific to remote sensing images by introducing a classification‐based self‐supervised auxiliary task. This auxiliary task involves performing image transformation self‐supervised learning tasks directly on the remote sensing images, thereby improving the overall capacity for feature representation. In this work, we design a texture fading reinforcement auxiliary task to reinforce texture features and color features that are useful for distinguishing similar classes of remote sensing. Different auxiliary tasks are fused to form a multi‐view self‐supervised auxiliary task and integrated with the main task to optimize the model training in an end‐to‐end manner. The experimental results on several popular few‐shot remote sensing image datasets validate the effectiveness of the proposed method. The performance better than many advanced algorithms is achieved with a more concise structure.

Computer-aided diagnosis of early-stage Retinopathy of Prematurity in neonatal fundus images using artificial intelligence

Retinopathy of Prematurity (ROP) is a retinal disorder affecting preterm babies, which can lead to permanent blindness without treatment. Early-stage ROP diagnosis is vital in providing optimal therapy for the neonates. The proposed study predicts early-stage ROP from neonatal fundus images using Machine Learning (ML) classifiers and Convolutional Neural Networks (CNN) based pre-trained networks. The characteristic demarcation lines and ridges in early stage ROP are segmented utilising a novel Swin U-Net. 2000 Scale Invariant Feature Transform (SIFT) descriptors were extracted from the segmented ridges and are dimensionally reduced to 50 features using Principal Component Analysis (PCA). Seven ROP-specific features, including six Gray Level Co-occurrence Matrix (GLCM) and ridge length features, are extracted from the segmented image and are fused with the PCA reduced 50 SIFT features. Finally, three ML classifiers, such as Support Vector Machine (SVM), Random Forest (RF), andk- Nearest Neighbor (k-NN), are used to classify the 50 features to predict the early-stage ROP from Normal images. On the other hand, the raw retinal images are classified directly into normal and early-stage ROP using six pre-trained classifiers, namely ResNet50, ShuffleNet V2, EfficientNet, MobileNet, VGG16, and DarkNet19. It is seen that the ResNet50 network outperformed all other networks in predicting early-stage ROP with 89.5% accuracy, 87.5% sensitivity, 91.5% specificity, 91.1% precision, 88% NPV and an Area Under the Curve (AUC) of 0.92. Swin U-Net Convolutional Neural Networks (CNN) segmented the ridges and demarcation lines with an accuracy of 89.7% with 80.5% precision, 92.6% recall, 75.76% IoU, and 0.86 as the Dice coefficient. The SVM classifier using the 57 features from the segmented images achieved a classification accuracy of 88.75%, sensitivity of 90%, specificity of 87.5%, and an AUC of 0.91. The system can be utilised as a point-of-care diagnostic tool for ROP diagnosis of neonates in remote areas.

The pilot study of the features of HIV-1 resistant variants spread using molecular clusters

Introduction. As a result of routine testing of HIV-1 drug resistance (DR), a significant amount of viral nucleotide sequences and epidemiological data of HIV-infected individuals have been collected. Combined with the increasing use of bioinformatics methods in practice, it has become possible to study the features of HIV-1 resistant variants spread using molecular clustering analysis. The aim of the study was to validate the molecular clustering analysis in a pilot region of Russia using a significant number of nucleotide sequences to study the features of the spread of HIV-1 resistant variants. Materials and methods. HIV-1 nucleotide sequences were obtained from 899 HIV-infected patients who were registered at the Oryol AIDS Center in 2016–2021. HIV-1 genetic variants were determined using the Stanford University database, REGA and HIV BLAST. Resistance mutations and prognostic HIV-1 DR were determined using the Stanford University database. Phylogenetic analysis was carried out using the MEGA program. HIV-1 molecular clusters were identified using Cluster Picker software. Results. In the pilot region, sub-subtype A6 dominated (85.7%); an increase in the share of CRF63_02A6 was noted. HIV-1 resistance was found in 13.6% of patients without antiretroviral therapy (ART) experience and in 52.0% with ART experience. Molecular clusters were more often formed by HIV-1 nucleotide sequences from ART-naïve patients. HIV-1 DR variants were less likely to fall into molecular clusters. The sources of transmitted mutations were more often patients with ART experience. The most actively and efficiently transmitted mutations were K103N, V179E/T, Y181C and G190S, associated with virus resistance to efavirenz and nevirapine.

Abstract A009: Integrative multiomic analysis of extracellular vesicle transcriptomics profiling combined with cfDNA methylation reveals improved stratification of low-risk and high- risk prostate cancer patients in urine-based liquid biopsy

Abstract Introduction: Prostate cancer (PCa) ranks as the second most frequently diagnosed cancer and the sixth most prevalent cause of cancer-related mortality in men globally. In this study, we employ a multiomic approach that utilizes both urine cfDNA and extracellular vesicle (EV) RNA-derived analytes to improve the risk stratification of individuals with prostate cancer and compare this to multi-parametric MRI performance as well as risk calculators in the same cohort. Materials and Methods: Urine was collected from 95 individuals comprising 51 benign or low-risk (Grade Group (GG) <=1), 10 intermediate-risk (GG2) and 44 high-risk PCa patients (GG 3-5). EV-RNA and cfDNA were concurrently isolated from each urine specimen to exploit complementary information. We developed an EV-RNA sequencing platform targeting mRNAs and lncRNAs and targeted 50 million reads per sample; and cfDNA methylome profiling reached an equivalent sequencing depth. Expression of EV-RNA and splice variant Differential Transcript Usage (DTU) and cfDNA methylation patterns were analyzed using Bio-Techne’s multiomic platform. Machine learning-based feature selection algorithms identified biomarker signatures from each analyte. Receiver-operator characteristic curves (ROC) were generated utilizing leave-one-out cross-validation of naïve Bayes classifier models to compute the area under the curve (AUC). Individual signatures were integrated to create a multiomic classifier. Results: Differential gene expression (DEx) analysis identified 18 DEx genes between low and high-risk patients with both known and novel implications in PCa. Splice variant analysis unveiled 60 genes previously implicated in PCa with DTU. Over 18,000 differentially methylated bases were detected between high and low-risk PCa patients. Feature selection analysis identified the top ten features of EV-RNA expression, splice variants, and cfDNA methylation, resulting in an integrative multi-analyte signature. The multiomic signature demonstrated an AUC of 0.92, surpassing any individual analyte’s performance, as well as the performance achieved by PSA (AUC = 0.52) and MRI PI RADS(AUC=0.65). Conclusions: While biomarker signatures obtained from each analyte in this study resulted in effective stratification of PCa risk, the multiomic signature further improved the discriminatory power—underscoring the complementary nature of the signals. Thus, a multiomic biomarker discovery strategy leveraging cfDNA and EV cargo exhibits significant potential as a risk assessment tool for high-grade prostate cancer. This approach can facilitate more informed decision-making in the context of biopsy procedures. Citation Format: Sudipto K Chakrabortty, Dulaney Miller, Kyle Manning, Rikky Xing, Jeff Cole, Christopher Benway, Sinead Nguyen, Allan George, Yiyuan Yao, Elena M Cortizas, Christian Ray, Siva Gowrisankar, Seth Yu, Allan Pollack, Sandra M Gaston, Sanoj Punnen, Johan Skog. Integrative multiomic analysis of extracellular vesicle transcriptomics profiling combined with cfDNA methylation reveals improved stratification of low-risk and high- risk prostate cancer patients in urine-based liquid biopsy [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr A009.

Fault diagnosis of rolling bearings driven by multi-channel data fusion and feature fusion under time-varying speed conditions

Abstract Bearings, as the core component for power transmission, are crucial in ensuring the safe and reliable operation of equipment. However, the fault information contained in a single-channel vibration signal is inherently limited. Additionally, under time-varying speed conditions, features are prone to drift, and the cross-domain diagnostic performance of most traditional domain adaptation (DA) models may drop dramatically. To solve the above problems and enhance the ability of DA models in extracting domain invariant features, this paper introduces a Multi-channel data fusion and Attention-guided Multi-feature Fusion-driven Center-aligned Network (MAMC). Initially, a multi-channel time-frequency information fusion strategy based on wavelet transform is constructed to achieve a comprehensive fusion of multi-channel data, thereby obtaining richer fault feature representations. Subsequently, a multi-branch feature fusion network, integrated with an attention mechanism, is devised to capture significant features across various dimensions and scales, resulting in more comprehensive and representative fault features. Finally, a novel Center-Aligned Domain Adaptation method (CADA) is proposed based on domain adversarial methods and center loss. By minimizing the distance between deep domain invariant features and trainable common class centers, the issue of domain shift between data is effectively alleviated, and the cross-domain diagnostic performance of DA models under the time-varying speed conditions is improved. The experimental results indicate that the MAMC method exhibits superior performance on both bearing datasets and is a promising approach for cross-domain intelligent fault diagnosis.

Primary Progressive Aphasia Lacking Core Features of Nonfluent and Semantic Variants

Primary Progressive Aphasia Lacking Core Features of Nonfluent and Semantic Variants

Zero-fault-shot learning for bearing spall type classification by hybrid approach

Spall type classification, a task in which the type of the spall is estimated, is an important stage for bearing diagnosis and prognosis. Many machine learning algorithms have been suggested for spall type classification. However, they are not relevant for diagnosing critical rotating machinery, where very few, if any, faulty examples (labeled and unlabeled) are available due to safety considerations. In this study, a novel hybrid algorithm is proposed, which enables the classification of the spall type based on zero-fault-shot learning. The novel algorithm combines physics-based algorithms together with machine learning to overcome the lack of faulty data. It projects the signals into an invariant feature space by physics-based algorithms and classifies the spall type by a fully connected neural network. The new algorithm is demonstrated on several well-known experimental datasets and significantly improves the performance of currently available learning algorithms for zero-fault-shot learning. It improves the results of the state-of-the-art algorithm from at most 60% on the six tested datasets to at least 98% accuracy on all tested datasets with the newly suggested algorithm.

Comprehensive empirical evaluation of feature extractors in computer vision

Feature detection and matching are fundamental components in computer vision, underpinning a broad spectrum of applications. This study offers a comprehensive evaluation of traditional feature detections and descriptors, analyzing methods such as Scale Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), Binary Robust Independent Elementary Features (BRIEF), Oriented FAST and Rotated BRIEF (ORB), Binary Robust Invariant Scalable Keypoints (BRISK), KAZE, Accelerated KAZE (AKAZE), Fast Retina Keypoint (FREAK), Dense and Accurate Invariant Scalable descriptor for Yale (DAISY), Features from Accelerated Segment Test (FAST), and STAR. Each feature extractor was assessed based on its architectural design and complexity, focusing on how these factors influence computational efficiency and robustness under various transformations. Utilizing the Image Matching Challenge Photo Tourism 2020 dataset, which includes over 1.5 million images, the study identifies the FAST algorithm as the most efficient detector when paired with the ORB descriptor and Brute-Force (BF) matcher, offering the fastest feature extraction and matching process. ORB is notably effective on affine-transformed and brightened images, while AKAZE excels in conditions involving blurring, fisheye distortion, image rotation, and perspective distortions. Through more than 2 million comparisons, the study highlights the feature extractors that demonstrate superior resilience across various conditions, including rotation, scaling, blurring, brightening, affine transformations, perspective distortions, fisheye distortion, and salt-and-pepper noise.

Recognition and classification of noise signals by dolphins under conditions of noise interference and spatial uncertainty of their simultaneous presentation

The ability of dolphin`s auditory system to recognize and classify noise signals according to certain invariant features under the influence of noise interference and spatial uncertainty of their simultaneous presentation was studied. The bottlenose dolphins trained to differentiate such signals had to solve this problem under conditions simulating real marine ones, when the perception of a useful signal occurs against the background of similar signals and noise interference. First, the noise signals were sequentially presented to the animal against the background of white masking noise. Further the dolphin had to identify a signal of a positive class from several simultaneously sounding sound sources. The efficiency of the animal was evaluated at several given levels of noise interference. In this case, the actual noise interference was both white noise and simultaneously sounding negative signals. It`s shown that the efficiency and noise immunity of dolphin`s auditory system depend on the degree of alternativeness of the spatial uncertainty of the simultaneous presentation of signals.

A Generalizable Causal-Invariance-Driven Segmentation Model for Peripancreatic Vessels.

Segmenting peripancreatic vessels in CT, including the superior mesenteric artery (SMA), the coeliac artery (CA), and the partial portal venous system (PPVS), is crucial for preoperative resectability analysis in pancreatic cancer. However, the clinical applicability of vessel segmentation methods is impeded by the low generalizability on multi-center data, mainly attributed to the wide variations in image appearance, namely the spurious correlation factor. Therefore, we propose a causal-invariance-driven generalizable segmentation model for peripancreatic vessels. It incorporates interventions at both image and feature levels to guide the model to capture causal information by enforcing consistency across datasets, thus enhancing the generalization performance. Specifically, firstly, a contrast-driven image intervention strategy is proposed to construct image-level interventions by generating images with various contrast-related appearances and seeking invariant causal features. Secondly, the feature intervention strategy is designed, where various patterns of feature bias across different centers are simulated to pursue invariant prediction. The proposed model achieved high DSC scores (79.69%, 82.62%, and 83.10%) for the three vessels on a cross-validation set containing 134 cases. Its generalizability was further confirmed on three independent test sets of 233 cases. Overall, the proposed method provides an accurate and generalizable segmentation model for peripancreatic vessels and offers a promising paradigm for increasing the generalizability of segmentation models from a causality perspective. Our source codes will be released at https://github.com/ SJTUBME-QianLab/PC_VesselSeg.

Multimodal invariant feature prompt network for brain tumor segmentation with missing modalities

Multimodal invariant feature prompt network for brain tumor segmentation with missing modalities

PCLC‐Net: Point Cloud Completion in Arbitrary Poses with Learnable Canonical Space

AbstractRecovering the complete structure from partial point clouds in arbitrary poses is challenging. Recently, many efforts have been made to address this problem by developing SO(3)‐equivariant completion networks or aligning the partial point clouds with a predefined canonical space before completion. However, these approaches are limited to random rotations only or demand costly pose annotation for model training. In this paper, we present a novel Network for Point cloud Completion with Learnable Canonical space (PCLC‐Net) to reduce the need for pose annotations and extract SE(3)‐invariant geometry features to improve the completion quality in arbitrary poses. Without pose annotations, our PCLC‐Net utilizes self‐supervised pose estimation to align the input partial point clouds to a canonical space that is learnable for an object category and subsequently performs shape completion in the learned canonical space. Our PCLC‐Net can complete partial point clouds with arbitrary SE(3) poses without requiring pose annotations for supervision. Our PCLC‐Net achieves state‐of‐the‐art results on shape completion with arbitrary SE(3) poses on both synthetic and real scanned data. To the best of our knowledge, our method is the first to achieve shape completion in arbitrary poses without pose annotations during network training.

Skin feature point tracking using deep feature encodings

AbstractFacial feature tracking is a key component of imaging ballistocardiography (BCG) where accurate quantification of the displacement of facial keypoints is needed for good heart rate estimation. Skin feature tracking enables video-based quantification of motor degradation in Parkinson’s disease. While traditional computer vision algorithms like Scale Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), and Lucas-Kanade method (LK) have been benchmarks due to their efficiency and accuracy, they often struggle with challenges like affine transformations and changes in illumination. In response, we propose a pipeline for feature tracking, that applies a convolutional stacked autoencoder to identify the most similar crop in an image to a reference crop containing the feature of interest. The autoencoder learns to represent image crops into deep feature encodings specific to the object category it is trained upon. We train the autoencoder on facial images and validate its ability to track skin features in general using manually labelled face and hand videos of small and large motion recorded in our lab. Our evaluation protocol is comprehensive, including quantification of errors in human annotation. The tracking errors of distinctive skin features (moles) are so small that we cannot exclude the fact that they stem from the manual labelling based on a $$\chi ^2$$ χ 2 -test. With a mean error of 0.6–3.3 pixels, our method outperformed the other methods in all but one scenario. More importantly, our method was the only one that did not diverge. We also compare our method with the latest state-of-the-art transformer for feature matching by Google—Omnimotion. Our results indicate that our method is superior at tracking different skin features under large motion conditions and that it creates better feature descriptors for tracking, matching, and image registration compared to both traditional algorithms and the latest Omnimotion.

Clinical characteristics of ABCB4 gene variant-associated cholestatic liver disease in adults

Objective: To investigate the clinical manifestations, pathological, and gene mutation characteristics of ABCB4 gene variant-associated cholestatic liver disease in adults. Methods: Eight adult cases of ABCB4 gene variant-associated cholestatic liver disease who were hospitalized in the Department of Hepatology, Fifth Medical Center of the People's Liberation Army General Hospital from May 2010 to December 2022 were enrolled in this study. The clinical manifestations, pathological features, gene variant features, and prognostic conditions were analyzed. Patient gene testing and biological information analysis were performed using whole-exome next-generation sequencing. SPSS 19.0 software was used to conduct descriptive analysis. Results: Among the eight adult cases of the ABCB4 gene variant, there were three males and five females, with a median age of onset of 24 (20, 37) years. There were three cases with a compound heterozygous variant in ABCB4, and the clinical phenotypes included two cases of progressive familial intrahepatic cholestasis type 3 and one case of intrahepatic cholestasis of pregnancy overlapping with low-phospholipid-associated cholelithiasis syndrome. There were five cases with a single heterozygous variant in ABCB4, and the clinical phenotypes included two cases of intrahepatic cholestasis of pregnancy overlapping with drug-induced liver injury and three cases of low-phospholipid-associated cholelithiasis syndrome. Imaging of all eight cases showed liver fibrosis, and six cases already had cirrhosis. All patients underwent liver histopathological examination, which mainly showed cholestasis and portal fibrosis in eight cases, small bile duct hyperplasia in seven cases, copper deposition in three cases, and cirrhosis in five cases. ABCB4 screening revealed 11 different mutations, including eight new mutations. The pathogenicity assessment showed that c.2394+82C>T (intron) was a benign mutation, and the rest were deleterious mutations. Ursodeoxycholic acid was the treatment for all patients, with a follow-up time of 7.5 (0.5, 12.7) years. One case died of end-stage liver disease, two cases developed cholestatic cirrhosis, and five cases were in stable condition. Conclusion: The adult ABCB4 gene variant-associated cholestatic liver disease are mostly single heterozygous mutations, the clinical phenotypes are diverse and overlapping, the disease is more severe in those who carried non-functional mutations.

MMCL: Meta-mutual contrastive learning for multi-modal medical image fusion

The number of datasets and computational efficiency are always hindrances in the multi-modal medical image fusion (MMIF) research. To address these challenges, we propose a contrastive learning framework inspired meta-mutual, which divides the medical image fusion task into subtasks and pre-trains an optimal meta-representation suitable for all subtasks. We then fine-tune our proposed network using this optimal meta-representation as initialization, achieving the best model with only a few short datasets. Additionally, extracting source image features in pairs can lead to redundant information due to the invariant and unique features of multi-modal images. Therefore, we introduce novelty mutual contrastive coupled pairs to extract both invariant and unique features from source images. Experimental results demonstrate that our method outperforms other state-of-the-art fusion methods.

Natural variability and individuality of walking behavior in Drosophila.

Insects use walking behavior in a large number of contexts, such as exploration, foraging, escape and pursuit, or migration. A lot is known on how nervous systems produce this behavior in general and also how certain parameters vary with regard to walking direction or speed, for instance. An aspect that has not received much attention is if and how walking behavior varies across individuals of a particular species. To address this, we created a large corpus of kinematic walking data of many individuals of the fruit fly Drosophila. We only selected instances of straight walking in a narrow range of walking speeds to minimize the influence of high-level parameters, like turning and walking speed, aiming to uncover more subtle aspects of variability. Using high-speed videography and automated annotation we captured the positions of the six leg tips for thousands of steps and used principal components analysis to characterize the postural space individuals used during walking. Our analysis shows that the largest part of walking kinematics can be described by five principal components (PCs). Separation of these five PCs into a 2-dimensional and a 3-dimensional subspace divided the description of walking behavior into invariant features shared across individuals and features that relate to the specifics of individuals; the latter features can be regarded as idiosyncrasies. We also demonstrate that this approach can detect the effects of experimental interventions in an unbiased manner and that general aspects of individuality, such as the the individual walking posture, can be described.

A validated heart-specific model for splice-disrupting variants in childhood heart disease

BackgroundCongenital heart disease (CHD) is the most common congenital anomaly. Almost 90% of isolated cases have an unexplained genetic etiology after clinical testing. Non-canonical splice variants that disrupt mRNA splicing through the loss or creation of exon boundaries are not routinely captured and/or evaluated by standard clinical genetic tests. Recent computational algorithms such as SpliceAI have shown an ability to predict such variants, but are not specific to cardiac-expressed genes and transcriptional isoforms.MethodsWe used genome sequencing (GS) (n = 1101 CHD probands) and myocardial RNA-Sequencing (RNA-Seq) (n = 154 CHD and n = 43 cardiomyopathy probands) to identify and validate splice disrupting variants, and to develop a heart-specific model for canonical and non-canonical splice variants that can be applied to patients with CHD and cardiomyopathy. Two thousand five hundred seventy GS samples from the Medical Genome Reference Bank were analyzed as healthy controls.ResultsOf 8583 rare DNA splice-disrupting variants initially identified using SpliceAI, 100 were associated with altered splice junctions in the corresponding patient myocardium affecting 95 genes. Using strength of myocardial gene expression and genome-wide DNA variant features that were confirmed to affect splicing in myocardial RNA, we trained a machine learning model for predicting cardiac-specific splice-disrupting variants (AUC 0.86 on internal validation). In a validation set of 48 CHD probands, the cardiac-specific model outperformed a SpliceAI model alone (AUC 0.94 vs 0.67 respectively). Application of this model to an additional 947 CHD probands with only GS data identified 1% patients with canonical and 11% patients with non-canonical splice-disrupting variants in CHD genes. Forty-nine percent of predicted splice-disrupting variants were intronic and > 10 bp from existing splice junctions. The burden of high-confidence splice-disrupting variants in CHD genes was 1.28-fold higher in CHD cases compared with healthy controls.ConclusionsA new cardiac-specific in silico model was developed using complementary GS and RNA-Seq data that improved genetic yield by identifying a significant burden of non-canonical splice variants associated with CHD that would not be detectable through panel or exome sequencing.

Fault domain separation networks: A cross-domain diagnostic method for variable operating conditions in the elevator guidance system

High-speed elevators play a critical role in vertical transportation. However, they may exhibit different working conditions under varying operating conditions, posing challenges for fault diagnosis. We propose a model named fault domain separation networks (FDSN) to tackle the cross-domain diagnostic problem of guidance system by leveraging transfer learning techniques. FDSN comprises three distinct modules that extract invariant fault features and variable operating condition features, and use orthogonal loss function for feature decoupling, achieving effective feature allocation, and mitigating the impact of variable operating conditions on fault diagnosis. To evaluate and demonstrate the efficacy of FDSN, a high-speed elevator platform is constructed, and fault vibration data under various working conditions are simulated to form a comprehensive dataset. The proposed FDSN achieved an average diagnostic accuracy of 92.7% in cross-domain diagnostic tasks, outperforming other comparative methods and demonstrating the remarkable superiority.

7962 Transformation of Poorly Differentiated Papillary Thyroid Carcinoma to Anaplastic Thyroid Carcinoma, an Atypical Course!

Abstract Disclosure: T. Altaweel: None. S. Makadsi: None. M. Hussein: None. J. Samantray: None. Introduction: Anaplastic thyroid cancers are undifferentiated, rare and aggressive. The transformation of papillary thyroid cancer to anaplastic thyroid cancer is documented in literature, however, rare. In this report, we present a case of Anaplastic thyroid carcinoma likely arising from a co-existing poorly differentiated Papillary carcinoma. Case Presentation: A 74-year-old male with past medical history of tobacco smoking who presented with the onset of voice hoarseness for 6 months. He has also noticed a left neck mass that is in the mid supraclavicular region. A Neck CT neck was performed at the time of presentation and was significant for a 4.6x3.4x5.9 cm left thyroid mass with displacement of the trachea. FNA biopsy showed papillary carcinoma with focal features of Tall Cell Variant. Poorly differentiated components with fragments of stroma and muscle were seen and suggested invasion. These findings raised the possibility of anaplastic thyroid carcinoma in the setting of papillary carcinoma and poorly differentiated carcinoma. BRAF V600E pathogenic variant mutation along with TERT mutation were found. Patient underwent direct laryngoscopy, bronchoscopy with biopsies, and tracheostomy with surgical pathology confirming papillary thyroid cancer with muscles invasion. Combination dabrafenib and trametinib were started. No surgical resection of thyroid tissue was planned. A follow up CT showed a significant decrease in the low-density soft tissue involving the medial thyroid gland bilaterally, consistent with interim positive response to therapy. Three months later, CT neck showed near complete resolution of low-density soft tissue involving the medial thyroid and parathyroid glands and subglottic trachea consistent with further interim positive response to therapy. Six months post initiation of therapy, CT Neck showed focal low density in the mid and inferior right thyroid gland which may represent residual and/or recurrent thyroid carcinoma. That density was further demonstrated on a repeat CT neck as a 1.8 cm hypodense nodule in the inferior right thyroid gland and concerning for recurrent thyroid tumor with possible infiltration of the trachea. Conclusion: ATC prognosis is affected by many factors such as metastasis, age, treatment strategies, socioeconomic status and the presence of BRAF V600E and TERT mutations. Diagnosis can be challenging. Our patient presented months after his symptoms started and this is an atypical course of Anaplastic Thyroid Carcinoma as it usually progresses aggressively and rapidly. Given the pathology reports and the more slow presentation of this patient’s carcinoma, it can be suggested that this is a case of papillary carcinoma with a later transformation to ATC. In these cases, surgical resection Presentation: 6/2/2024

Test time adaptation for cross-domain sleep stage classification

Domain shift is a serious problem with existing cross-domain sleep stage classification methods. However, most existing domain adaptation methods require access to a large amount of source domain data containing patient information, which may lead to infringement of patient privacy. At the same time, existing methods require collecting all information in advance when processing target data, which cannot meet the situation where test data arrives in batches or in order in actual scenarios. To this end, we propose a test time adaptation framework based on knowledge distillation for cross-domain sleep stage classification. This teacher-student network framework uses pre-trained model to learn variant features and robust representations. We evaluate our framework in the cross domain scenario composed of three datasets and validate its superiority, real time performance and privacy protection capability.